Abstract
We report a web-based tool for analysis of experiments using indirect calorimetry to measure physiological energy balance. CalR simplifies the process to import raw data files, generate plots, and determine the most appropriate statistical tests for interpretation. Analysis using the generalized linear model (which includes ANOVA and ANCOVA) allows for flexibility in interpreting diverse experimental designs, including those of obesity and thermogenesis. Users also may produce standardized output files for an experiment that can be shared and subsequently re-evaluated using CalR. This framework will provide the transparency necessary to enhance consistency, rigor, and reproducibility. The CalR analysis software will greatly increase the speed and efficiency with which metabolic experiments can be organized, analyzed per accepted norms, and reproduced and has become a standard tool for the field. CalR is accessible at https://CalRapp.org/
The top 4 key questions that our tool can answer:
1. Can I reproducibly and transparently analyze indirect calorimetry experiments in under 10 minutes?
2. How hard is it to use Analysis of Covariance (ANCOVA) to determine whether my groups of animals are significantly different?
3. Is there an automated, reproducible way to exclude “noisy” outlier data from our indirect calorimetry experiments?
4. What are the key factors in determining metabolic rate of mice?
Presenter: Alexander Banks, PhD, principal investigator and assistant professor at Harvard Medical School and the Beth Israel Deaconess Medical Center.
dkNET Webinar Information: https://dknet.org/about/webinar
Pests of mustard_Identification_Management_Dr.UPR.pdf
dkNET Webinar: A New Approach to the Study of Energy Balance and Obesity using CalR 05/08/2020
1. A New Approach to the Study of Energy
Balance and Obesity
Alex Banks, Ph.D.
Assistant Professor
Beth Israel Deaconess Medical Center
Director, Energy Balance Core
Harvard Medical School
2. Talk Overview
• CalR: A statistical tool for metabolic scientists
• Live data analysis using CalR
3. Using Calorimetry to Measure Energy Balance
The 1st law of thermodynamics, conservation of energy.
Calories In = Energy Expenditure + Calories not absorbed
Getty images
= +
Neutral Energy Balance
Indirect Calorimeter (measures food intake and
calculates energy expenditure from gas exchange)
Direct calorimeter (measures
energy retained in fecal matter
through combustion)
4. Indirect calorimetry on two or four legs
Humans (maximum metabolic rate) Humans (steady state metabolic rate) Julia Belluz, Vox Sept 2018 Mice (steady state metabolic rate)
Indirect calorimetry is a calculated estimate of energy expenditure
using experimentally determined formulae
EE (kcal/time) = 3.88 VO2 + 1.08 VCO2 - 1.57 N
EE (kcal/time) = 3.85 VO2 + 1.07 VCO2
assuming similar protein content in diets and weight stability, can be simplified:
5. Metabolic rate of mammals scales with body mass: Klieber (1932)
The single greatest source of calculation error in obesity studies
The Problem:
Bigger animals have greater metabolic requirements
To compare diverse species of different sizes you can adjust for the larger
body size by the Klieber scaling factor
Metabolic rate Body mass 2/3 for masses < 10 kg
or
Metabolic rate Body mass 3/4 for masses > 10 kg
However, obese mice can’t be compared to lean mice by Klieber scaling:
Klieber scaling assumes similar body compositions
BodyComposition(g)
other
fat
lean
fat mass is 80-85% less metabolically
active compared to lean mass
Mammals and birds
6. Metabolic Rate (normalized by total body mass)
Pelleymounter M.A., et al. Science 1995
10 mg/kg
leptin
A teachable moment in how to analyze obesity studies correctly:
Why do leptin-deficient ob/ob mice lose weight when treated with leptin?
Is energy expenditure affected? Food intake?
Himms-Hagen J. Science 1997
ob/ob body weight (% change)
10 mg/kg leptin
PBS
PBS
10 mg/kg leptin
ob/ob food intake (% change), not normalized
Artifacts are introduced by weight-
based normalization
foodintake(%D)bodyweight(%D)
Days Kaiyala et al, Molecular Metabolism 2016
7. What is the right way to examine these experiments?
The ‘39 steps’: an algorithm for performing statistical
analysis of data on energy intake and expenditure
(with scripts in Minitab, R, SPSS)
John R. Speakman, Quinn Fletcher, Lobke Vaanholt;
Disease Models & Mechanisms 2013
8. Yet these problems continue to recur
• When body composition is different, dividing by total mass is an invalid method of analysis.
• Complete time periods should be compared, not individual time points.
• Investigators and students need help with data analysis
This is still being done in 2020!
Kraakman MJ et al, Journal of Clinical Investigation 2018
Zhao P et al, Cell 2018
Choi MJ et al. Diabetologia 2020
9. How should we interpret this plot?
Oxygenconsumption
(ml/hr)
Oxygenconsumption
(ml/kg/hr)
Oxygenconsumption
(ml/kg/hr)
Oxygenconsumption
(ml/kg/hr)
Both groups have the same metabolic
rate when body mass is the covariate.
“Mass Effect”
One group has a greater metabolic
rate when body mass is the
covariate.
“Group Effect”
One group with obese mice has
a lower metabolic rate when lean
body mass is the covariate.
“Group Effect”
Oxygenconsumption
(ml/hr)
Oxygenconsumption
(ml/hr)
Mass
ANCOVA = not significant
Oxygenconsumption
(ml/hr)
Mass
ANCOVA = significant
Mass
ANCOVA = significant
11. What is a “normal” mouse metabolic rate?
The effects of mass, temperature, and sex
Sex Temperature
Relative contribution to
variation in energy
expenditure
n = 3,164 wildtype C57Bl/6 mice; Age 10-11 weeks, chow diet, data from the International Mouse Phenotyping Consortium (IMPC) Corrigan et al. eLife May 2020
12. Statistical approach for experiments of obesity and
thermogenesis
Mina et al. Cell Metabolism 2018
13. The original approach:
>8 hours of analysis with Microsoft Excel per experiment
Or a custom Python, SAS or R script for each experiment
14. CalR: A Web Tool for Analysis of Indirect
Calorimetry and Energy Expenditure
CalR
Calor. Latin: heat
&
designed in the R statistical
programming language
With Louise Lantier and David E. Cohen
Mina et al. Cell Metabolism 2018
CalRapp.org/
15. How do we define an experiment?
As simply as possible.
Key compromise: one indirect calorimetry run may be multiple “experiments”
(e.g. 30C vs 4C ). They should be analyzed separately in CalR.
30 C 4 C
6 Analysis templates to accommodate common experimental designs
16. What kind of experiments can
we analyze with CalR?
CalR can analyze the most
common experimental
designs and has the
appropriate statistical
frameworks on different web
pages
CalR cannot analyze all
experimental designs
2 groups
2 groups:
acute response
< 12 hrs
3 groups:
Ordered groups
3 groups:
Not-ordered groups
4 groups:
Not-ordered groups
17. Graphical User Interface: Loading Data
✔️
✔️
✔️
Disconnected from the server.
If CalR cannot read your file, it gives this error:
18. Graphical User Interface: Visualizing Results
Credit for usability of the graphical user interface: Louise Lantier at Vanderbilt
19. Analysis with ANCOVA and GLM
For each GLM/ANCOVA, does greater mass correlate with each variable across all animals in the study
Mass-dependent variables only
Mass-independent variables only
Are the groups significantly different?
20. Understanding the mass effect and
the group effect
Sex Temperature
n = 3,164 wildtype C57Bl/6 mice; Age 10-11 weeks, chow diet, data from the International Mouse Phenotyping Consortium (IMPC) Corrigan et al. eLife May 2020
A significant mass effect:
The mass of the animal affects the metabolic variable
(a bigger mouse has a larger EE than a small mouse).
A significant mass effect is a good indicator of a well-
functioning experimental system.
A significant group effect:
The two groups are significantly different when
accounting for differences in mass.
Investigators are usually interested in whether their
group effect is significant.
*
21. Real-world data can have real problems (1/2):
Excluding data after animals are removed
groupindividual
Remove excluded mouse
4C cold challenge
22. Real-world data can have real problems (2/2):
optional automatic outlier detection (±3 SD)
groupindividual
Remove outliers turned on (±3 standard deviations per group per photoperiod)
23. Summary of the CalR Project:
A statistical tool for metabolic scientists
Simplify the analysis of energy balance experiments using indirect calorimetry
• Pre-define statistical approach and cutoffs
• Remove statistical decisions from the hands of end users
• Increase reproducibility, transparency and rigor
• Read and standardize data from all manufacturers of indirect calorimetery systems
• Clean data sets (they are large and noisy)
• Increase speed of analysis
• Promote collaboration and sharing of data with a standardized file format & compatibility
• Anonymous (no experimental data is stored)
• (Future) provide a repository for indirect calorimetry data as reference dataset
CalRapp.org/
A free open-source web tool where investigators can upload their raw data,
perform quality control, perform the appropriate multiple regression statistical
analysis, and generate publication-quality graphs and tables.
Current (May 1, 2020) usage statistics:
12,798 sessions, 4,227 users (75% USA), 33 countries
Avg Session Duration 7min 21 sec
-
2,000
4,000
6,000
8,000
10,000
12,000
Sep-17
Nov-17
Dec-17
Feb-18
Apr-18
May-18
Jul-18
Sep-18
Oct-18
Dec-18
Feb-19
Mar-19
May-19
Jul-19
Aug-19
Oct-19
Dec-19
Jan-20
Mar-20
Apr-20
CALR TOTAL SESSIONS
(AVERAGE SESSION TIME 7:21)
Cumulative
Sessions
24. Example 1: What is the source of excess body weight in diet-
induced obesity?
Corrigan et al. eLife May 2020n=8 per diet WT C57Bl/6J mice from JAX, UC Davis MMPC
26. Despite significant development of obesity on HFD, no
significant differences between groups of energy expenditure,
intake or balance
energyexpenditure(kcal/hr)
energybalance(kcal/hr)
energyintake(kcal/hr)
body mass (g) body mass (g)body mass (g)
energy expenditure energy balanceenergy intake
n=8 per diet C57Bl/6J mice from JAX, UC Davis MMPC
ns
ns ns
27. Number of mice needed to detect significant differences
Fernández-Verdejo, et al Nature Methods 2019
28. Summary of Example 1:
• WT mice on HFD gain more weight due to fat mass (15 g vs 5g)
• By indirect calorimetry, we detect no significant differences in food intake or energy
expenditure. However, a trend toward increased food and decreased expenditure gives a
non-significant trend for increased energy balance.
• Possibilities include unaccounted for difference in energy absorption.
• The most likely solution is that this effect size is small and noisy, but still biologically
relevant. If EE were the main driver of these weight differences, we would need more than
400 mice to detect a significant result.
• Could the classic p<0.05 be too conservative for studies of energy balance?
29. Example 2: Two groups with large effect size differences
30. CalR data analysis work flow
CalR data file: Original, unaltered raw data from experiment
CalR session file: metadata including
• Subject IDs
• Group names, group color assignments
• Assignment of subjects to groups
• Subject body weights or body compositions
• Specify photoperiod
• Specify diet energy content
• Selected time range for analysis
• Exclude ±3SD outliers per photoperiod?
• Subject exclusion?
31. What is the physiological relevance of any given significant difference in
energy expenditure?
What is “normal”?
Energy Expenditure
WT
body mass
WT
Strain 1
Strain 2
Strain 3
Strain 4
Strain 5
32. 15 w old
94 w old
3 w exercise
0 w exercise
CL at 30°C
30°C
6°C
10°C
14°C
22°C
25°C
28°C
55 w old
1 w exercise
2 w exercise
18°C
30°C
Normal mouse physiology: in context
33. 21 34 170
1821
126 41 33
IMPC: distribution of EE phenotypes in 2600 KO strains
35. Best practices for analysis of obesity & energy
balance experiments
• Show regression plots to see mass dependence
• Analysis by ANCOVA/GLM
• Analyze groups before body weight (or body composition) diverges
• Use body composition data (MRI or DEXA) to increase precision
• Never ever plot: VO2 (ml/kg/hr) or VO2(ml/kg0.75/hr)
• Share your data in a repository or as supplemental data
CalR is a simple solution to this problem and may not fit all complex scenarios
Yao L et al, Elife 2018 Kazak et al. Cell Metabolism 2017
36. Acknowledgements
Banks Lab at Beth Israel Deconess
Amir Mina (CalR)
June Corrigan (Large-scale analysis)
Deepti Ramachandran
Yuchen He
Katie LeClaire
Critical Collaborators
Louise Lantier @ Vanderbilt
Owen McGuinness @ Vanderbilt
David E. Cohen @ Weill Cornell
Ray LeClaire @ Springbok
John Lighton @ Sable Systems
Harry Knot @ TSE
MMPC Energy Balance Working Group
Funding:
NIDDK National Mouse Metabolic Phenotyping Centers (Micromouse Grant)
Harvard Digestive Disease Centers
NIH S10 Instrumentation Grant (Indirect Calorimeter)
NIDDK R01
Amir Mina
(now MD/PhD student,
University of Pittsburgh)
June Corrigan